Messages in this thread Patch in this message | | | Date | Tue, 2 Jul 2013 10:52:07 +0800 | Subject | Re: [V2 2/2] sched: update cfs_rq weight earlier in enqueue_entity | From | Lei Wen <> |
| |
Paul,
On Mon, Jul 1, 2013 at 10:07 PM, Paul Turner <pjt@google.com> wrote: > Could you please restate the below? > > On Mon, Jul 1, 2013 at 5:33 AM, Lei Wen <leiwen@marvell.com> wrote: >> Since we are going to calculate cfs_rq's average ratio by >> runnable_load_avg/load.weight > > I don't understand what you mean by this.
Previously I take runnable_load_avg/load.weight calculation as the cfs_rq's average ratio. But as Alex point out, the runnable_avg_sum/runnable_avg_period may better sever this need.
> >>, if not increase the load.weight prior to >> enqueue_entity_load_avg, it may lead to one cfs_rq's avg ratio higher >> than 100%. >> > > Or this.
In my mind, runnable_load_avg in one cfs_rq should always be less than load.weight. Not sure whether this assumption stand here, but runnable_load_avg/load.weight truly could shows out the cfs_rq execution trend in some aspect.
The previous problem that enqueue_entity_load_avg called before account_entity_enqueue, which make runnable_load_avg be updated first, then the load.weight. So that with the trace info log inside of enqueue_entity_load_avg, we may get the calculation result for runnable_load_avg/load.weight > 1. This result is not friendly for the final data being parsed out.
> >> Adjust the sequence, so that all ratio is kept below 100%. >> >> Signed-off-by: Lei Wen <leiwen@marvell.com> >> --- >> kernel/sched/fair.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c >> index 07bd74c..d1eee84 100644 >> --- a/kernel/sched/fair.c >> +++ b/kernel/sched/fair.c >> @@ -1788,8 +1788,8 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) >> * Update run-time statistics of the 'current'. >> */ >> update_curr(cfs_rq); >> - enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP); >> account_entity_enqueue(cfs_rq, se); >> + enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP); > > account_entity_enqueue is independent of enqueue_entity_load_avg; > their order should not matter.
Yes, agree, the order should not be matter, but for make trace info integrated, we may need some order here.
> > Further, should we restore the reverted amortization commit (improves > context switch times)
Not understand here... What the "should we restore the reverted amortization commit (improves context switch times)" means here...?
enqueue_entity_load_avg needs to precede > account_entity_enqueue as it may update se->load.weight.
account_entity_enqueue needs to precede enqueue_entity_load_avg?
Thanks, Lei
> >> update_cfs_shares(cfs_rq); >> >> if (flags & ENQUEUE_WAKEUP) { >> -- >> 1.7.10.4 >> > -- > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/
| |